EEG-Based Stress Detection: EEMD-Enhanced Features with SVM and Twin-SVM Models

Recent advances in signal processing and machine learning have significantly advanced EEG-based applications, especially in emotion detection and cognitive analysis. Using Intrinsic Mode Functions (IMFs) developed from Ensemble Empirical Mode Decomposition (EEMD) for improved feature extraction, this work investigates an EEG-based emotional stress identification system. The EEG signals were collected in EDF format, pre-processed by segmenting them into 10-second chunks, ensuring consistent and manageable data windows across all channels. IMFs were extracted from these segments to capture complex frequency components, which were then used as input features for machine learning classifiers. Support Vector Machine (SVM) and Twin Support Vector Machine (Twin SVM) classifiers were used to assess the system. The SVM classifier achieved an accuracy of 80.75%, while the Twin SVM classifier demonstrated a higher training accuracy of 89. 75% and Twin SVM shows an improvement in test accuracy of 93.75%. These results suggest that, while SVM offers good training performance, Twin SVM provides more reliable generalization to unseen data. The findings offer valuable insight into effective pre-processing techniques and the potential of these classifiers for real-time stress detection systems.